A Machine Learning Approach for Identifying Favorable Sites for Renewable Energy Installations

This paper demonstrates the application of machine learning in determining suitability for utility scale sites for renewable energy production. Supervised algorithms such as Random Forest Classifier are employed in a Semi-Supervised learning process that allows underlying trends present in suitable and non suitable sites to be extrapolated to a mostly unlabeled dataset. The model iteratively trains from the pseudo labels it creates throughout this process, until all data points in the data set are labeled. This allows the small percentage of hand labeled data to be leveraged for use in the larger dataset. This open source tool can be used by anyone for the quick and precise determination of suitable locations for utility scale and personal installations based on the available renewable resources. It can also be used to influence future policy decisions around renewable energy.